Electric vehicles still represent less than 1% of registered vehicles in the U.S. and comprised just over 8% of new U.S. vehicle sales in 2024. But government officials across the country are already planning for a future in which EVs are the go-to mode of private transportation — as evidenced by increasingly large public investments in EV charging infrastructure, like the $7.5 billion pledged under the Biden administration. When it comes to actually making a plan for building out that infrastructure, however, the devil is in a complex web of largely unknown details, says Associate Professor of Industrial and Manufacturing Systems Engineering Jian Hu. Ideally, Hu says, you’d want to build a charging system that serves people’s needs for the next 10 years. The tricky part is that involves correctly anticipating a host of trendlines which are basically impossible to predict. “There are so many uncertainties,” Hu says. “We don’t know how fast the EV market will grow, which means we don’t know what the charging demand will be. Will the public want fast or slow charging? Will people charge at home or at work? During peak hours, or non-peak hours? Will utilities adopt dynamic pricing so it’s cheaper to charge at certain times of day? And how will this affect charging behavior? And what if there are new technologies that emerge? There are so many things we don’t know.”
Typically, when government agencies are planning for things like this, Hu says they might use what statisticians call a deterministic model. When you’re making this kind of model, you take account of what data you have, plug it into a formula that defines the relationship between the variables, and then you typically get a single answer, which then guides your plan of action. A simple retirement calculator, which takes your current age, retirement age, anticipated rate of return, the rate of inflation, and monthly savings target and tells you how much money you’ll have when you retire is a simple kind of deterministic model. But Hu argues that these kinds of prediction models aren’t the best tools when you’re dealing with complex situations with lots of uncertainty. “What if your rate of return on your investments isn’t the 7% you’ve put into that calculator? What if it is less and then you don’t have the money you expect when you retire? If you only use this kind of model, you’ll have this single prediction and you base everything on that. But you’re not preparing for a possible worst-case scenario.”
Because of this, statisticians often use different tools for forecasting complex, highly uncertain phenomena, like investment performance. One of the most popular is called stochastic modeling, which, instead of giving you just one answer, provides a range of results represented by thousands of different simulated scenarios in which the values and probabilities of the underlying variables are allowed to change. A popular type of stochastic model is a Monte Carlo simulation, which is used in financial planning. Unlike a simple retirement calculator, which assumes fixed values for things like the rate of inflation or a rate of return, a Monte Carlo analysis will run thousands of simulations in which all the variables you have in your model have different values and assigned probabilities. As a result, you get a much wider representation of possible outcomes, including best-case, worst-case and most likely scenarios. A Monte Carlo simulation might tell you, say, that in 98 percent of its simulated scenarios, which account for a wide range of possible economic conditions, your savings plans are going to get you to the end of your life with money in the bank. That might give you much more confidence in your investment strategy. You might even decide to take a more aggressive approach. Or, as Hu frames it, stochastic modeling gives us a much more complete way to understand risk.
Because stochastic models are so much better at helping us understand risk of uncertain events, Hu says they could be an excellent approach for planning EV infrastructure, given that reliable data on important details, like consumer charging preferences and future electricity demand, are sparse or lacking altogether. As part of a National Science Foundation-funded project, and in collaboration with Argonne National Laboratory, which is itself partnering with the Chicago Transit Authority, Hu will be creating novel approaches to stochastic modeling that he hopes can profoundly improve this planning process. To get just a little bit technical for a moment: Hu says one of the big challenges of stochastic modeling is coping with “distributional ambiguity,” which is where you’re lacking a lot of reliable information about the probability of key phenomena. Currently, one of the ways statisticians deal with this is a method called distributionally robust optimization, but Hu says DRO has a few limitations. One, it’s computationally intensive, which means as models get more complex, it becomes increasingly cumbersome and time consuming to run them and adapt them to new data. Second, DRO tends to hedge against worst-case scenarios and can therefore lead to overly conservative outcomes. Third, it's a machine learning-based process that is a “black box,” meaning it does not show how it arrives at its answers. It therefore might not be a great option for publicly funded agencies, who, in spending large sums of taxpayer dollars, could certainly benefit from transparent or interpretable models. Hu says his novel stochastic modeling strategy aims not only to improve on these limitations, but also provide more overall usability for public agencies, including models that can be more easily updated as better data emerge.
After developing the model, Hu is looking forward to a big field test toward the end of the project: helping the Chicago Transit Authority as it embarks on electrifying their massive fleet of public buses. Current electric bus battery technology generally provides a range of around 60 miles. But many of CTA’s current scheduled vehicle route blocks are much longer; nearly a third of weekday blocks are longer than 100 miles. CTA therefore envisions building out a bus-charging infrastructure that’s integrated with their garages and bus routes, which will also require enhancements to the local electric grid to accommodate the new electricity load. Figuring out where these grid modifications need to be made and the ideal placement for the charging stations in a way that harmonizes with the complex bus routes — which are likely to change in the future — is exactly the kind of complicated, ambiguous planning task Hu says his model can help with. Most importantly, it gives a government agency its best chance of spending limited public resources for EV charging as effectively as possible.
###
Story by Lou Blouin
